The STIS/CCD is a low-noise device capable of high sensitivity in the visible and the near UV. It is a thinned, backside-illuminated device manufactured by Scientific Imaging Technologies (SITe). In order to provide a near-UV imaging performance, the CCD was backside-treated and coated with a wide-band anti-reflectance coating. The process produces acceptable near-UV quantum efficiency (QE) without compromising the high QE of the visible bandpass. The CCD camera design incorporates a warm dewar window, designed to prevent buildup of contaminants on the window, which were found to cause a loss of UV throughput for the WFPC2 CCDs. A summary of the STIS CCD performance is given in Table 7.1. The performance values on read noise and dark current are those valid as of February 2001.
The spectral response of the unfiltered CCD is shown in Figure 5.1 (labeled as 50CCD). This figure illustrates the extremely wide bandpass over which this CCD can operate. The wide wavelength coverage is an advantage for deep optical imaging. The near-UV sensitivity of the CCD makes it a good alternative to the NUV-MAMA for low- and intermediate-resolution spectroscopy from ~2500 to 3100 Å using the G230LB and G230MB grating modes ( Table 4.1).
Based on data to date, the STIS CCD does not suffer from Quantum Efficiency Hysteresis (QEH)-that is, the CCD responds in the same way to light levels over its whole dynamic range, irrespective of the previous illumination level.
Monitoring of the CCD sensitivity through October 2000 shows mode-dependent trends with time. The G430L and G750L modes show no variation, but the G230LB shows a decrease of approximately 1.5% per year. Tests of selected medium resolution modes show a range of decrease in average sensitivity from roughly 1% per year for the G230M, G430M and G750M modes, to almost 2% per year for the G230MB mode. A more complete description of the change in sensitivity over time can be found in STIS ISR 2000-03.
Figure 7.1 shows the change in relative sensitivity of the G430M setting versus time, together with a linear fit to the points. The value for the slope, i.e., the percentage change in sensitivity per year, and the 1
uncertainty are shown at the bottom of the plot. The 1 rms (in %) of the data about the fit is indicated by the value in the same plot.
Like most CCDs, the STIS CCD exhibits fringing in the red, longward of ~7500 Å. This fringing limits the signal-to-noise routinely achievable in the red and near-IR unless contemporaneous flats are obtained (see below). In principle, fringing can also affect imaging observations if the source's emission over the 50CCD or F28X50LP bandpass is dominated by emission lines redward of 7500 Å. However, if the bulk of the emission comes from blueward of 7500 Å, then emission from multiple wavelengths will smooth over the fringe pattern so that imaging will not be affected by fringing.
The amplitude of the fringes is a strong function of wavelength and spectral resolution. Table 7.2 lists the observed percentile peak-to-peak and rms amplitudes of the fringes as a function of central wavelength for the G750M and G750L gratings. The listed "peak-to-peak" amplitudes are the best measure of the impact of the fringing on your data. The rms values at wavelengths < 7000 Å give a good indication of the counting statistics in the flat-field images used for this analysis.
The fringe pattern can be corrected for by rectification with an appropriate flat field. The fringe pattern is a convolution of the contours of constant distance between the front and back surfaces of the CCD and the wavelength of the light on a particular part of the CCD. The fringe pattern has been shown to be very stable, as long as the wavelength of light on a particular part of the CCD stays constant. However, due to the grating wheel positioning uncertainty ( Slit and Grating Wheels) and the effect of temperature drifts in orbit, the wavelength on a particular part of the CCD will vary from observation to observation. Thus, the best de-fringing results are obtained by using a contemporaneous flat ("fringe flat"), i.e., a tungsten flat taken at the same grating wheel setting, during the same orbit as your scientific exposures.
Table 7.3 compares the estimated peak-to-peak fringe amplitudes after flat-fielding by the library flat and by those after flat fielding with an appropriately processed contemporaneous flat. These estimates are based upon actual measurements of spectra of both point sources and extended sources, made during Cycle 7 (the results for point sources and extended sources were essentially the same). Figure 7.2 shows such a comparison for a G750L spectrum of a white dwarf; in this figure, the top panel shows white dwarf GD153 (central wavelength 7751 Å) with no flat field correction, the second spectrum shows the result of de-fringing with the standard pipeline flat field, and the third spectrum shows the result of de-fringing with a contemporaneous flat (all spectra were divided by a smooth spline fit to the stellar continuum). It is clear that a contemporaneous flat provides a great improvement over the use of a library flat. Therefore, if you are observing in the far red (> 7500 Å) and using grating G750L or G750M, you should take a contemporaneous flat field along with your scientific observations. More detailed information and analysis on fringe correction for STIS long-wavelength spectra can be found in STIS ISRs 98-19, 98-29, and the references therein.
| Wavelength (Å) |
G750M, library flat: residual |
G750M, contemp. flat: residual1 |
G750L, library flat: residual |
G750L, contemp. flat: residual |
|---|---|---|---|---|
| 7500 |
3.0 |
1.2 |
4.0 |
0.9 |
| 7750 |
2.5 |
1.3 |
5.3 |
0.8 |
| 8000 |
4.2 |
1.3 |
7.5 |
1.0 |
| 8250 |
4 |
1.0 |
5.3 |
0.9 |
| 8500 |
5 |
0.9 |
8.3 |
1.0 |
| 8750 |
6 |
0.9 |
6.5 |
0.9 |
| 9000 |
8 |
- |
8.3 |
1.0 |
| 9250 |
10 |
- |
17.5 |
1.4 |
| 9500 |
11 |
- |
18.7 |
2.0 |
| 9750 |
12 |
- |
19.7 |
2.4 |
| 10000 |
10 |
- |
11.7 |
2.4 |
1 Measurements of the fringe amplitude have not been made yet for G750M wavelength settings redward of 8561 Å. However, from our experience with fringe corrections we expect the residual fringe amplitudes to be of order 1% when contemporaneous fringe flats are used. |
Examination of long slit observations in the CCD spectroscopic modes has revealed periodic variations of intensity along the slit when highly monochromatic, calibration lamp sources are used. An example of such `chevron-pattern' variations is shown in Figure 7.3 and Figure 7.4. These variations are thought to be the result of transmission variations through the highly parallel faces of the order sorting filters, situated next to the grating in the optical path. In the cross dispersion direction, the modulation amplitude depends on the line width with a maximum of 13% for a monochromatic source in G430L and G430M modes, and 4.5% in G750L and G750M modes. Periods range from 40 - 80 pixels/cycle. In the dispersion direction, there is a small, residual, high frequency modulation with a peak amplitude of about 1.5% in G430M at the 5471Å setting and with smaller amplitudes in all other modes and settings. No such modulation has been observed in any of the MAMA modes. Some modulation is also apparent in the CCD G230LB and G230MB modes, however the amplitudes of these modulations are much smaller and analysis of these modes is ongoing.
It should be noted that this effect is pronounced only for monochromatic sources; the modulation is negligible for continuum sources. So far, such modulation has not been observed in any astronomical observations. A calibration program would be necessary to investigate this effect further.
Figure 7.4: A plot of the 52x0.2 slit illumination near 4269Å in mode G430L obtained with the Pt-Cr/Ne calibration lamp
Verification testing has shown that STIS meets its image-quality specifications. While the optics provide fine images at the focal plane, the detected point-spread functions (PSFs) are degraded somewhat more than expected by the CCD at wavelengths longward of about 7500 Å, where a broad halo appears, surrounding the PSF core. This halo is believed to be due to scatter within the CCD mounting substrate, which becomes more pronounced as the silicon transparency increases at long wavelengths. The effects of the red halo, which extend to radii greater than 100 pixels (5 arcsec), are not included in the encircled energies as a function of observing wavelength which are described for the CCD spectroscopic modes and the CCD imaging modes in Chapter 13 and Chapter 14, respectively. The integrated energy in the halo amounts to approximately 20% of the total at 8050 Å and 30% at 9050 Å (see also STIS ISR 97-13 for the implication for long-slit spectroscopic observations at long wavelengths).
The CCD plate scale is 0.05071 arcseconds per pixel, for imaging observations and in the spatial (across the dispersion) direction for spectroscopic observations. Due to the effect of anamorphic magnification, the plate scale in the dispersion direction is slightly different and dependent on the grating used and its tilt. The plate scale in the dispersion direction ranges from 0.05121 to 0.05727 arcseconds per pixel (see STIS ISR 98-20).
The CCD detector produces a relatively faint, out-of-focus, ring-shaped "ghost" image, due to specular reflection from the CCD surface and window. The ring contains about 1% of the total energy in the image and is very stable. Additional rings of similar size can be seen at other locations in the field in grossly saturated images, but these contain only of order 10-5 of the total energy and are thus not likely to be detected in normal scientific images. Lines drawn from stars in images through their respective ghosts are found to converge at a "radiant point" located to the lower right of the image center. This effect is illustrated in Figure 7.5 where the line segments are drawn from pixel coordinates 528,342 (in 1024 x 1024 user coordinates) through the centroids of the brightest stars in the image. Note that these line segments intercept the centers of the ring-like ghosts very well. Observers who wish to avoid placing very faint objects within the range of the ghosts may want to take this geometry into account when writing Phase II submissions.
Figure 7.5: Ring-Shaped Ghost Images Near Bright Point Sources (50CCD Image)
A full detector readout is 1062 x 1044 pixels including physical and virtual overscans. Scientific data are obtained on 1024 x 1024 pixels, each projecting to ~0.05 x 0.05 arcseconds on the sky. For spectroscopic observations, the dispersion axis runs along axis1 (image X or along a row of the CCD), and the spatial axis of the slits runs along axis2 (image Y or along a column of the CCD). The CCD supports the use of subarrays to read out only a portion of the detector, and on-chip binning. For more details see "CCD ACCUM Mode" on page 212.
Electrons which accumulate in the CCD wells are read out and converted to data numbers (DN, the format of the output image) by the analog-to-digital converter at a default CCDGAIN of 1 e-/DN (i.e., every electron registers 1 DN). The CCD is also capable of operating at a gain of 4 e-/DN. The analog-to-digital converter operates at 16 bits, producing a maximum of 65,536 DN pixel-1. This is not a limitation at either gain setting, since other factors set the maximum observable DN to lower levels in each case ( CCD Operation and Feasibility Considerations below).
The CCDGAIN=1 setting has the lower readout noise ( Table 7.1) and digitization noise, and is thus the most appropriate setting for observations of faint sources. Unfortunately however, saturation is already occurring at about 33,000 e- at the CCDGAIN=1 setting (as described in CCD Operation and Feasibility Considerations below). Greater accumulations, up to the CCD full well limit of 144,000 e-, can be observed at CCDGAIN=4.
While the CCDGAIN=4 setting does allow one to reach the CCD full well, short exposures taken in CCDGAIN=4 show large-scale pattern noise ("ripple") that is not removed by the standard bias images. Figure 7.6 (a 0.2 second exposure of a lamp-illuminated small slit) shows an example of this effect. The peak-to-peak intensities of the ripples vary from near zero to about 1 DN, and there is a large amount of coherence in the noise pattern. This coherence makes background determination difficult and limits the precision of photometry in shallow exposures taken in CCDGAIN=4. Use of the CCDGAIN=4 setting for imaging photometry is recommended for photometry of objects based on large counts (>104 e-).
|
Space Telescope Science Institute http://www.stsci.edu Voice: (410) 338-1082 help@stsci.edu |